X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=src%2Fbin%2Flttng-sessiond%2Frotation-thread.c;h=8e63e160f980c15f8e60af0ea8f74d88a89e758e;hp=5b0267952676ccc549a95c7a3b21bab3a70d413b;hb=9c9d917cfe2e6a9f35ef82cb64e442451210716f;hpb=92816cc33a1add3c8276839bd6335e17423577dd

diff --git a/src/bin/lttng-sessiond/rotation-thread.c b/src/bin/lttng-sessiond/rotation-thread.c
index 5b0267952..8e63e160f 100644
--- a/src/bin/lttng-sessiond/rotation-thread.c
+++ b/src/bin/lttng-sessiond/rotation-thread.c
@@ -42,8 +42,10 @@
 #include "rotate.h"
 #include "cmd.h"
 #include "session.h"
-#include "sessiond-timer.h"
+#include "timer.h"
 #include "notification-thread-commands.h"
+#include "utils.h"
+#include "thread.h"
 
 #include <urcu.h>
 #include <urcu/list.h>
@@ -56,7 +58,7 @@ struct rotation_thread {
 
 struct rotation_thread_job {
 	enum rotation_thread_job_type type;
-	uint64_t session_id;
+	struct ltt_session *session;
 	/* List member in struct rotation_thread_timer_queue. */
 	struct cds_list_head head;
 };
@@ -72,11 +74,11 @@ struct rotation_thread_timer_queue {
 };
 
 struct rotation_thread_handle {
-	int quit_pipe;
 	struct rotation_thread_timer_queue *rotation_timer_queue;
 	/* Access to the notification thread cmd_queue */
 	struct notification_thread_handle *notification_thread_handle;
-	sem_t *notification_thread_ready;
+	/* Thread-specific quit pipe. */
+	struct lttng_pipe *quit_pipe;
 };
 
 static
@@ -132,15 +134,13 @@ void log_job_destruction(const struct rotation_thread_job *job)
 		abort();
 	}
 
-	LOG(log_level, "Rotation thread timer queue still contains job of type %s targeting session %" PRIu64 " on destruction",
-			job_type_str, job->session_id);
+	LOG(log_level, "Rotation thread timer queue still contains job of type %s targeting session \"%s\" on destruction",
+			job_type_str, job->session->name);
 }
 
 void rotation_thread_timer_queue_destroy(
 		struct rotation_thread_timer_queue *queue)
 {
-	struct rotation_thread_job *job, *tmp_job;
-
 	if (!queue) {
 		return;
 	}
@@ -148,12 +148,7 @@ void rotation_thread_timer_queue_destroy(
 	lttng_pipe_destroy(queue->event_pipe);
 
 	pthread_mutex_lock(&queue->lock);
-	/* Empty wait queue. */
-	cds_list_for_each_entry_safe(job, tmp_job, &queue->list, head) {
-		log_job_destruction(job);
-		cds_list_del(&job->head);
-		free(job);
-	}
+	assert(cds_list_empty(&queue->list));
 	pthread_mutex_unlock(&queue->lock);
 	pthread_mutex_destroy(&queue->lock);
 	free(queue);
@@ -165,14 +160,13 @@ void rotation_thread_timer_queue_destroy(
 void rotation_thread_handle_destroy(
 		struct rotation_thread_handle *handle)
 {
+	lttng_pipe_destroy(handle->quit_pipe);
 	free(handle);
 }
 
 struct rotation_thread_handle *rotation_thread_handle_create(
-		int quit_pipe,
 		struct rotation_thread_timer_queue *rotation_timer_queue,
-		struct notification_thread_handle *notification_thread_handle,
-		sem_t *notification_thread_ready)
+		struct notification_thread_handle *notification_thread_handle)
 {
 	struct rotation_thread_handle *handle;
 
@@ -181,13 +175,18 @@ struct rotation_thread_handle *rotation_thread_handle_create(
 		goto end;
 	}
 
-	handle->quit_pipe = quit_pipe;
 	handle->rotation_timer_queue = rotation_timer_queue;
 	handle->notification_thread_handle = notification_thread_handle;
-	handle->notification_thread_ready = notification_thread_ready;
+	handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
+	if (!handle->quit_pipe) {
+		goto error;
+	}
 
 end:
 	return handle;
+error:
+	rotation_thread_handle_destroy(handle);
+	return NULL;
 }
 
 /*
@@ -196,13 +195,14 @@ end:
  */
 static
 bool timer_job_exists(const struct rotation_thread_timer_queue *queue,
-		enum rotation_thread_job_type job_type, uint64_t session_id)
+		enum rotation_thread_job_type job_type,
+		struct ltt_session *session)
 {
 	bool exists = false;
 	struct rotation_thread_job *job;
 
 	cds_list_for_each_entry(job, &queue->list, head) {
-		if (job->session_id == session_id && job->type == job_type) {
+		if (job->session == session && job->type == job_type) {
 			exists = true;
 			goto end;
 		}
@@ -212,7 +212,8 @@ end:
 }
 
 void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
-		enum rotation_thread_job_type job_type, uint64_t session_id)
+		enum rotation_thread_job_type job_type,
+		struct ltt_session *session)
 {
 	int ret;
 	const char * const dummy = "!";
@@ -220,7 +221,7 @@ void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
 	const char *job_type_str = get_job_type_str(job_type);
 
 	pthread_mutex_lock(&queue->lock);
-	if (timer_job_exists(queue, session_id, job_type)) {
+	if (timer_job_exists(queue, job_type, session)) {
 		/*
 		 * This timer job is already pending, we don't need to add
 		 * it.
@@ -230,12 +231,15 @@ void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
 
 	job = zmalloc(sizeof(struct rotation_thread_job));
 	if (!job) {
-		PERROR("Failed to allocate rotation thread job of type \"%s\" for session id %" PRIu64,
-				job_type_str, session_id);
+		PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"",
+				job_type_str, session->name);
 		goto end;
 	}
+	/* No reason for this to fail as the caller must hold a reference. */
+	(void) session_get(session);
+
+	job->session = session;
 	job->type = job_type;
-	job->session_id = session_id;
 	cds_list_add_tail(&job->head, &queue->list);
 
 	ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), dummy,
@@ -256,8 +260,8 @@ void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
 			DBG("Wake-up pipe of rotation thread job queue is full");
 			goto end;
 		}
-		PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session id %" PRIu64,
-				job_type_str, session_id);
+		PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"",
+				job_type_str, session->name);
 		goto end;
 	}
 
@@ -272,30 +276,32 @@ int init_poll_set(struct lttng_poll_event *poll_set,
 	int ret;
 
 	/*
-	 * Create pollset with size 2:
-	 *	- quit pipe,
+	 * Create pollset with size 3:
+	 *	- rotation thread quit pipe,
 	 *	- rotation thread timer queue pipe,
+	 *	- notification channel sock,
 	 */
-	ret = lttng_poll_create(poll_set, 2, LTTNG_CLOEXEC);
+	ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC);
 	if (ret < 0) {
-		goto end;
+		goto error;
 	}
 
-	ret = lttng_poll_add(poll_set, handle->quit_pipe,
+	ret = lttng_poll_add(poll_set,
+			lttng_pipe_get_readfd(handle->quit_pipe),
 			LPOLLIN | LPOLLERR);
 	if (ret < 0) {
-		ERR("[rotation-thread] Failed to add quit_pipe fd to pollset");
+		ERR("[rotation-thread] Failed to add quit pipe read fd to poll set");
 		goto error;
 	}
+
 	ret = lttng_poll_add(poll_set,
 			lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe),
 			LPOLLIN | LPOLLERR);
 	if (ret < 0) {
-		ERR("[rotation-thread] Failed to add rotate_pending fd to pollset");
+		ERR("[rotation-thread] Failed to add rotate_pending fd to poll set");
 		goto error;
 	}
 
-end:
 	return ret;
 error:
 	lttng_poll_clean(poll_set);
@@ -326,11 +332,6 @@ int init_thread_state(struct rotation_thread_handle *handle,
 		goto end;
 	}
 
-	/*
-	 * We wait until the notification thread is ready to create the
-	 * notification channel and add it to the poll_set.
-	 */
-	sem_wait(handle->notification_thread_ready);
 	rotate_notification_channel = lttng_notification_channel_create(
 			lttng_session_daemon_notification_endpoint);
 	if (!rotate_notification_channel) {
@@ -394,7 +395,7 @@ int check_session_rotation_pending_local_on_consumer(
 static
 int check_session_rotation_pending_local(struct ltt_session *session)
 {
-	int ret;
+	int ret = 0;
 	struct consumer_socket *socket;
 	struct cds_lfht_iter iter;
 	bool rotation_completed = true;
@@ -442,7 +443,12 @@ end:
 		session->rotation_pending_local = false;
 	}
 	if (ret) {
-		session->rotation_state = LTTNG_ROTATION_STATE_ERROR;
+		ret = session_reset_rotation_state(session,
+				LTTNG_ROTATION_STATE_ERROR);
+		if (ret) {
+			ERR("Failed to reset rotation state of session \"%s\"",
+					session->name);
+		}
 	}
 	return 0;
 }
@@ -502,14 +508,19 @@ int check_session_rotation_pending_relay(struct ltt_session *session)
 		ERR("[rotation-thread] Encountered an error when checking if rotation of trace archive %" PRIu64 " of session \"%s\" is pending on the relay",
 				session->current_archive_id - 1,
 				session->name);
-		session->rotation_state = LTTNG_ROTATION_STATE_ERROR;
+		ret = session_reset_rotation_state(session,
+				LTTNG_ROTATION_STATE_ERROR);
+		if (ret) {
+			ERR("Failed to reset rotation state of session \"%s\"",
+					session->name);
+		}
 		rotation_completed = false;
 	}
 
 	rcu_read_unlock();
 
 	if (rotation_completed) {
-		DBG("[rotation-thread] Totation of trace archive %" PRIu64 " of session \"%s\" is complete on the relay",
+		DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on the relay",
 				session->current_archive_id - 1,
 				session->name);
 		session->rotation_pending_relay = false;
@@ -531,6 +542,19 @@ int check_session_rotation_pending(struct ltt_session *session,
 	DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64,
 			session->name, session->current_archive_id - 1);
 
+	/*
+	 * The rotation-pending check timer of a session is launched in
+	 * one-shot mode. If the rotation is incomplete, the rotation
+	 * thread will re-enable the pending-check timer.
+	 *
+	 * The timer thread can't stop the timer itself since it is involved
+	 * in the check for the timer's quiescence.
+	 */
+	ret = timer_session_rotation_pending_check_stop(session);
+	if (ret) {
+		goto end;
+	}
+
 	if (session->rotation_pending_local) {
 		/* Updates session->rotation_pending_local as needed. */
 		ret = check_session_rotation_pending_local(session);
@@ -566,7 +590,12 @@ int check_session_rotation_pending(struct ltt_session *session,
 	/* Rename the completed trace archive's location. */
 	now = time(NULL);
 	if (now == (time_t) -1) {
-		session->rotation_state = LTTNG_ROTATION_STATE_ERROR;
+		ret = session_reset_rotation_state(session,
+				LTTNG_ROTATION_STATE_ERROR);
+		if (ret) {
+			ERR("Failed to reset rotation state of session \"%s\"",
+					session->name);
+		}
 		ret = LTTNG_ERR_UNK;
 		goto end;
 	}
@@ -598,6 +627,44 @@ int check_session_rotation_pending(struct ltt_session *session,
 				session->name);
 	}
 
+	if (!session->active) {
+		/*
+		 * A stop command was issued during the rotation, it is
+		 * up to the rotation completion check to perform the
+		 * renaming of the last chunk that was produced.
+		 */
+		ret = notification_thread_command_session_rotation_ongoing(
+				notification_thread_handle,
+				session->name,
+				session->uid,
+				session->gid,
+				session->current_archive_id);
+		if (ret != LTTNG_OK) {
+			ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
+					session->name);
+		}
+
+		ret = rename_active_chunk(session);
+		if (ret < 0) {
+			ERR("[rotation-thread] Failed to rename active rotation chunk");
+			goto end;
+		}
+
+		/* Ownership of location is transferred. */
+		location = session_get_trace_archive_location(session);
+		ret = notification_thread_command_session_rotation_completed(
+				notification_thread_handle,
+				session->name,
+				session->uid,
+				session->gid,
+				session->current_archive_id,
+				location);
+		if (ret != LTTNG_OK) {
+			ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
+					session->name);
+		}
+	}
+
 	ret = 0;
 end:
 	if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) {
@@ -615,7 +682,7 @@ end:
 	return ret;
 }
 
-/* Call with the session lock held. */
+/* Call with the session and session_list locks held. */
 static
 int launch_session_rotation(struct ltt_session *session)
 {
@@ -663,18 +730,9 @@ int handle_job_queue(struct rotation_thread_handle *handle,
 		struct rotation_thread_timer_queue *queue)
 {
 	int ret = 0;
-	int fd = lttng_pipe_get_readfd(queue->event_pipe);
-	struct ltt_session *session;
-	char buf;
-
-	ret = lttng_read(fd, &buf, 1);
-	if (ret != 1) {
-		ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd);
-		ret = -1;
-		goto end;
-	}
 
 	for (;;) {
+		struct ltt_session *session;
 		struct rotation_thread_job *job;
 
 		/* Take the queue lock only to pop an element from the list. */
@@ -689,10 +747,10 @@ int handle_job_queue(struct rotation_thread_handle *handle,
 		pthread_mutex_unlock(&queue->lock);
 
 		session_lock_list();
-		session = session_find_by_id(job->session_id);
+		session = job->session;
 		if (!session) {
-			DBG("[rotation-thread] Session %" PRIu64 " not found",
-					job->session_id);
+			DBG("[rotation-thread] Session \"%s\" not found",
+					session->name);
 			/*
 			 * This is a non-fatal error, and we cannot report it to
 			 * the user (timer), so just print the error and
@@ -705,14 +763,16 @@ int handle_job_queue(struct rotation_thread_handle *handle,
 			 */
 			session_unlock_list();
 			free(job);
+			session_put(session);
 			continue;
 		}
 
 		session_lock(session);
-		session_unlock_list();
-
 	        ret = run_job(job, session, handle->notification_thread_handle);
 		session_unlock(session);
+		/* Release reference held by the job. */
+		session_put(session);
+		session_unlock_list();
 		free(job);
 		if (ret) {
 			goto end;
@@ -777,7 +837,7 @@ int handle_condition(const struct lttng_condition *condition,
 	ret = unsubscribe_session_consumed_size_rotation(session,
 			notification_thread_handle);
 	if (ret) {
-		goto end;
+		goto end_unlock;
 	}
 
 	ret = cmd_rotate_session(session, NULL);
@@ -800,6 +860,7 @@ int handle_condition(const struct lttng_condition *condition,
 
 end_unlock:
 	session_unlock(session);
+	session_put(session);
 end:
 	return ret;
 }
@@ -819,7 +880,7 @@ int handle_notification_channel(int fd,
 	status = lttng_notification_channel_has_pending_notification(
 			rotate_notification_channel, &notification_pending);
 	if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) {
-		ERR("[rotation-thread ]Error occured while checking for pending notification");
+		ERR("[rotation-thread ]Error occurred while checking for pending notification");
 		ret = -1;
 		goto end;
 	}
@@ -868,6 +929,8 @@ void *thread_rotation(void *data)
 	int ret;
 	struct rotation_thread_handle *handle = data;
 	struct rotation_thread thread;
+	const int queue_pipe_fd = lttng_pipe_get_readfd(
+			handle->rotation_timer_queue->event_pipe);
 
 	DBG("[rotation-thread] Started rotation thread");
 
@@ -876,17 +939,17 @@ void *thread_rotation(void *data)
 		goto end;
 	}
 
+	rcu_register_thread();
+	rcu_thread_online();
+
 	health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION);
 	health_code_update();
 
 	ret = init_thread_state(handle, &thread);
 	if (ret) {
-		goto end;
+		goto error;
 	}
 
-	/* Ready to handle client connections. */
-	sessiond_notify_ready();
-
 	while (true) {
 		int fd_count, i;
 
@@ -919,23 +982,41 @@ void *thread_rotation(void *data)
 				goto error;
 			}
 
-			if (fd == handle->quit_pipe) {
-				DBG("[rotation-thread] Quit pipe activity");
-				/* TODO flush the queue. */
-				goto exit;
-			} else if (fd == lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe)) {
+			if (fd == rotate_notification_channel->socket) {
+				ret = handle_notification_channel(fd, handle,
+						&thread);
+				if (ret) {
+					ERR("[rotation-thread] Error occurred while handling activity on notification channel socket");
+					goto error;
+				}
+			} else {
+				/* Job queue or quit pipe activity. */
+
+				/*
+				 * The job queue is serviced if there is
+				 * activity on the quit pipe to ensure it is
+				 * flushed and all references held in the queue
+				 * are released.
+				 */
 				ret = handle_job_queue(handle, &thread,
 						handle->rotation_timer_queue);
 				if (ret) {
 					ERR("[rotation-thread] Failed to handle rotation timer pipe event");
 					goto error;
 				}
-			} else if (fd == rotate_notification_channel->socket) {
-				ret = handle_notification_channel(fd, handle,
-						&thread);
-				if (ret) {
-					ERR("[rotation-thread] Error occured while handling activity on notification channel socket");
-					goto error;
+
+				if (fd == queue_pipe_fd) {
+					char buf;
+
+					ret = lttng_read(fd, &buf, 1);
+					if (ret != 1) {
+						ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd);
+						ret = -1;
+						goto error;
+					}
+				} else {
+					DBG("[rotation-thread] Quit pipe activity");
+					goto exit;
 				}
 			}
 		}
@@ -945,6 +1026,35 @@ error:
 	DBG("[rotation-thread] Exit");
 	fini_thread_state(&thread);
 	health_unregister(health_sessiond);
+	rcu_thread_offline();
+	rcu_unregister_thread();
 end:
 	return NULL;
 }
+
+static
+bool shutdown_rotation_thread(void *thread_data)
+{
+	struct rotation_thread_handle *handle = thread_data;
+	const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe);
+
+	return notify_thread_pipe(write_fd) == 1;
+}
+
+bool launch_rotation_thread(struct rotation_thread_handle *handle)
+{
+	struct lttng_thread *thread;
+
+	thread = lttng_thread_create("Rotation",
+			thread_rotation,
+			shutdown_rotation_thread,
+			NULL,
+			handle);
+	if (!thread) {
+		goto error;
+	}
+	lttng_thread_put(thread);
+	return true;
+error:
+	return false;
+}